ITERATE-2.V: scheduler priority aging closes 18-day AUDIT-049 wedge
Priority aging in xenia-cpu/scheduler.rs:pick_runnable
(effective_priority = base + age_bonus(now_round - last_run_round),
capped at +31, AGING_ROUNDS_PER_BONUS=1). Strict-priority was parking
priority=0 threads behind CPU-bound priority=15 audio mixer
(sub_824D1328 guest spinwait at PC=0x824d1404 on CPU5). Aging
eventually picks the starved thread, breaking the producer-consumer
cycle that caused 5-tid wedge at PC=0x824ac578 since AUDIT-049 (10 May).
Cascade observed: tid=13 clean exit; events 121K -> 13M (107x); last
host_ns 767ms -> 51,011ms (66x); 8 new threads spawn; VdSwap 1 -> 2.
Complete two-day iterate sequence (2026-05-27 -> 2026-05-28):
- 2.F: VdSwap drain timeout 900ms -> 1ms (xenia-gpu/handle.rs); 876x
perf win on VdSwap kernel callback
- 2.H: vA0000000 physical heap bucket added (state.rs, exports.rs);
ctx_ptrs now in 0xA0000000-0xBFFFFFFF range matching canary
- 2.L: Phase-A diff harness categorized [return_value mismatch],
[status mismatch], [args_resolved.path mismatch] tags
(tools/diff-events/diff_events.py); closes reading-error #41
(silent test-harness state leak invalidating trace diffs)
- 2.M: always-on exit-thread-state.json sibling to Phase-A JSONL
(event_log.rs + xenia-app/main.rs); closes reading-error #42
(Phase-A blind to blocked-forever waits)
- 2.Q: signal.match kernel instrumentation in NtSetEvent /
NtReleaseSemaphore / KeSetEvent / KeReleaseSemaphore
(exports.rs); emits target_handle + waiter_count + waiter_tids
- 2.T: wake.requested kernel instrumentation in wake_eligible_waiters
(exports.rs); emits target_tid + transition + new_state
- 2.V: scheduler priority aging (xenia-cpu/scheduler.rs) [keystone]
Plus accumulated WIP from earlier May (contention_manifest,
phase_b_snapshot, xam/xaudio enhancements, analysis db, xex loader,
xenia-app main loop, etc.). Audit-runs/ artifacts remain untracked
per project convention.
Tests: 300 xenia-cpu / 227 xenia-kernel / 5 xenia-app / 19 xenia-path
/ 30+ smaller suites -- all PASS, 0 regressions. Determinism preserved
(2x cold runs bit-identical at 13,003,881 events post-2.V).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
MechaCat02
56
docs/functions/sub_824ACB38.md
Normal file
56
docs/functions/sub_824ACB38.md
Normal file
@@ -0,0 +1,56 @@
|
||||
---
|
||||
address: 0x824ACB38
|
||||
classification: crt_init_driver
|
||||
confidence: high
|
||||
last_audit: 060
|
||||
aliases:
|
||||
- "CRT driver"
|
||||
- "vtable-slot enumerator (NOT a static-ctor list iterator)"
|
||||
---
|
||||
|
||||
# sub_824ACB38 — CRT init driver / vtable-slot enumerator
|
||||
|
||||
## Synopsis
|
||||
|
||||
CRT-style driver called from `entry_point` (or near it). Body is 224 bytes (`0x824ACB38..0x824ACC18`). Contains two enumeration loops over fnptr-array regions at `0x82870xxx`. AUDIT-050 framed it as "iterates 0x82870xxx fnptr arrays (557 slots, 82 non-NULL)" and concluded a half-bootstrapped state; AUDIT-060 found this framing semantically misleading — the slots are runtime vtable-registration entries, not C++ static initializers, and the "82 non-NULL" count obscures a structural 160-slot intentional zero gap.
|
||||
|
||||
## Evidence
|
||||
|
||||
- Body anatomy at `0x824ACB38..0x824ACC18` (AUDIT-060 disasm):
|
||||
- `+0x00..+0x2C` — preamble + one optional dispatch through fn-ptr at `[0x82023F08]` (= `0x825F1630`, an LZ-runtime thunk).
|
||||
- `+0x30..+0x6C` — **loop A**: enumerate u32 slots in `[0x828708C8, 0x828708D4)` — 3 slots. Filter: non-NULL. `bctrl` at `0x824ACBA0`.
|
||||
- `+0x80..+0xB8` — **loop B**: enumerate u32 slots in `[0x82870010, 0x828708C4)` — 557 slots. Filter: non-NULL AND `!= 0xFFFFFFFF`. `bctrl` at `0x824ACBEC`.
|
||||
- `+0xC4` — epilogue, `blr`.
|
||||
- Array layout (AUDIT-060 dumped at -n 1M and -n 500M; both identical):
|
||||
- `0x82870010..0x828702E8` — populated with `0x82xxxxxx` pointers (vtable methods).
|
||||
- `0x828702F0..0x82870580` — **PERMANENTLY ZERO** across both dumps (160 of 557 slots = 28.7%).
|
||||
- `0x82870590..0x828708C4` — populated with `0x82xxxxxx` pointers.
|
||||
- `0x828708C8..0x828708D4` — loop-A array, populated (small CRT helpers).
|
||||
- Static-DB cross-check: the 557-slot region contains 14+ separate small `vtable`-classified arrays at `0x82870014/0x24/0x94/0xC8/0x16C/0x214/0x238/0x250/0x2A8/0x2C0/0x2E4/0x5A0/0x62C/0x870`, NOT a single CRT static-ctor list. NO statically-detected arrays in `[0x82870300, 0x828705A0)` — the gap is intentional padding between two vtable clusters.
|
||||
|
||||
## Activation
|
||||
|
||||
Called once from `entry_point`-near code (per AUDIT-050 — exact caller PC not in AUDIT-060 trace). The driver enumerates all slot entries; each non-NULL entry is `bctrl`'d once.
|
||||
|
||||
## Static graph
|
||||
|
||||
- Static callers: 1 (from boot entry path; exact PC to confirm).
|
||||
- Callees: indirect (`bctrl`) — targets are the contents of the enumerated slots.
|
||||
|
||||
## Audit log
|
||||
|
||||
- **AUDIT-060 (2026-05-12)** — disassembled body, identified structure (2 loops, gap), confirmed slot contents are runtime vtable entries rather than C++ static-ctor function pointers. The "82 non-NULL" AUDIT-050 count was correct per-slot but missed the structural 160-slot intentional gap. Driver fires 1× at -n 500M as expected (single boot enumeration). [confirmed]
|
||||
- **AUDIT-050 (2026-05-10)** — framed as "CRT driver iterates 0x82870xxx fnptr arrays (557 slots, 82 non-NULL)". Structurally correct but semantically misleading ("static-ctor list" implication was wrong). [STATUS: partially superseded by AUDIT-060 — the "iterate fnptr array" claim stands; the "static-ctor list" implication does not.]
|
||||
|
||||
## Open questions
|
||||
|
||||
- What invokes this driver in `entry_point`? Find exact caller PC.
|
||||
- Are the 14+ small vtable clusters in `[0x82870010, 0x828708C4)` enumerated by THIS driver, or by separate driver functions? If multiple drivers exist for the same region, the gap might be region-partitioning, not padding.
|
||||
- For ours: are all 397 non-NULL slots dispatched at runtime? If some slot dispatch falls through (e.g. predicate skips it), that would be a real bug — needs runtime confirmation via `--branch-probe=0x824ACBA0,0x824ACBEC` (loop bodies).
|
||||
|
||||
## Cross-references
|
||||
|
||||
- LZ-runtime thunk at `+0x00..+0x2C`: `0x825F1630`.
|
||||
- Fnptr-array region: `0x82870010..0x828708D4`.
|
||||
- Audits: 045 (DB schema caveat: `v_call_graph` uses `xrefs.source`; prefer `xrefs.source_func`), 050, 060.
|
||||
- Artifacts: `audit-runs/audit-060-fnptr-array-bootstrap/ours-dump-500M.stdout`.
|
||||
Reference in New Issue
Block a user